MA 110:

Lecture 03

Saurav Bhaumik
Department of Mathematics
IIT Bombay

Spring 2025

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

Suppose an m×n matrix A is in a REF, and there are r
nonzero rows. Let the r pivots be in the columns k1 , . . . , kr
with k1 < · · · < kr , and let the columns ℓ1 , . . . , ℓn−r be
nonpivotal. Then xk1 , . . . , xkr are called the pivotal
variables and xℓ1 , . . . , xℓn−r are called the free variables.
Let b := [b1 · · · br br +1 · · · bm ]T , and consider the linear system
Ax = b.
Important Observations
1. The linear system has a solution ⇐⇒ br +1 = · · · = bm = 0.
This is known as the consistency condition.
2. Let the consistency condition br +1 = · · · = bm = 0 be
satisfied. Then we obtain a particular solution
 T
x0 := x1 · · · xn of the linear system by letting xk := 0
if k ∈ {ℓ1 , . . . , ℓn−r }, and then by determining the pivotal
variables xk1 , . . . , xkr by back substitution.
Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03
3. We obtain n − r basic solutions of the homogeneous
linear system Ax = 0 as follows. Fix ℓ ∈ {ℓ1 , . . . , ℓn−r }.
 T
Define sℓ := x1 · · · xn by xk := 1 if k = ℓ, while
xk := 0 if k ∈ {ℓ1 , . . . , ℓn−r } but k ̸= ℓ. Then determine
the pivotal variables xk1 , . . . , xkr by back substitution.
 T
4. Let s := x1 · · · xn ∈ Rn×1 be any solution of the
homogeneous system, that is, As = 0. Then s is a linear
combination of the n − r basic solutions sℓ1 , . . . , sℓn−r .
To see this, let y := s − xℓ1 sℓ1 − · · · − xℓn−r sℓn−r . Then
Ay = As − xℓ1 Asℓ1 − · · · − xℓn−r Asℓn−r = 0, and moreover,
the kth entry of y is 0 for each k ∈ {ℓ1 , . . . , ℓn−r }. It then
follows that y = 0, that is, s = xℓ1 sℓ1 + · · · + xℓn−r sℓn−r .
Thus we find that the general solution of the
homogeneous system is given by
s = α1 sℓ1 + · · · + αn−r sℓn−r , where α1 , . . . , αn−r ∈ R .

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

 5. The general solution of Ax = b is given by

x = x0 + α1 sℓ1 + · · · + αn−r sℓn−r , where α1 , . . . , αn−r ∈ R,

provided the consistency condition is satisfied.

Example
   
0 2 1 0 2 5 0
1 .
0 0 0 3 5 0  
Let A := 
0 and b := 2
0 0 0 1 2
0 0 0 0 0 0 0

As we have seen earlier, here m = 4, n = 6, r = 3, pivotal

columns: 2 , 4 and 5 , and nonpivotal columns: 1 , 3 , 6 .
Since b4 = 0, the linear system Ax = b is consistent.

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

For a particular solution of Ax = b, let x1 = x3 = x6 = 0.
Then x5 + 2x6 = 2 =⇒ x5 = 2,
3x4 + 5x5 + 0x6 = 1 =⇒ x4 = −3,
2x2 + x3 + 0x
 4 + 2x5 + 5x6 = 0 =⇒T x2 = −2.
Thus x0 := 0 −2 0 −3 2 0 is a particular solution.
Basic solutions of Ax = 0:  T
x1 = 1, x3 = x6 = 0 gives s1 := 1 0 0 0 0 0 ,
 T
x3 = 1, x1 = x6 = 0 gives s3 := 0 −1/2 1 0 0 0 ,
 T
x6 = 1, x1 = x3 = 0 gives s6 := 0 −1/2 0 10/3 −2 1 .
The general solution of Ax = b is given by
x = x0 + α1 s1 + α3 s3 + α6 s6 , that is,
x1 = α1 , x2 = −2 − (α3 + α6 )/2, x3 = α3 , x4 = −3 + 10α6 /3,
x5 = 2(1 − α6 ), x6 = α6 , where α1 , α3 , α6 ∈ R.

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

Conclusion
Suppose an m × n matrix A is in a REF, and let r be the
number of nonzero rows of A. If b ∈ Rm×1 , then the linear
system Ax = b has
(i) no solution if one of br +1 , . . . , bm is nonzero.
(ii) a unique solution if br +1 = · · · = bm = 0 and r = n.
(iii) infinitely many solutions if br +1 = · · · = bm = 0 and
r < n. In this case, there are n − r free variables which give


n − r degrees of freedom .
Considering the case b = 0 ∈ Rm×1 and recalling that r ≤ m,
we obtain the following important results.
Proposition
Let A ∈ Rm×n be in REF with r nonzero rows. Then the linear
system Ax = 0 has only the zero solution if and only if r = n.
In particular, if m < n, then Ax = 0 has a nonzero solution.
Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03
Gauss Elimination Method (GEM)

We have seen how to solve the linear system Ax = b when the

matrix A is in a row echelon form (REF).
We now explain the Gauss Elimination Method (GEM) by
which we can transform any A ∈ Rm×n to a REF.
This involves the following two elementary row operations
(EROs):

Type I: Interchange of two rows

Type II: Addition of a scalar multiple of a row to another row

We shall later consider the following elementary row operation:

Type III: Multiplication of a row by a nonzero scalar

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

Consider the matrix Ti,j , which is obtained from the identity
matrix by interchanging the row i and the row j. Then
1
 ..
.

 
0 1
 
 

Ti,j =  . ..


 

 1 0 

 . .

 . 
1
Then for any matrix A, the result of interchanging the row i
and row j of A is given by the matrix Ti,j A.

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

For any constant α, consider the matrix Li,j (α), which is
obtained from the identity
 matrix by adding α in the  (i, j)-th
1
 ..
.

 
1 α
 
 

position. So, Lij (α) =  . .

 . 


 1 

 . .

 . 
1
If A is a matrix, then the result of adding α times the row j to
the row i is given by the matrix Li,j (α)A.

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

First we remark that if the augmented matrix [A|b] is
transformed to a matrix [A′ |b′ ] by any of the EROs, then
Ax = b ⇐⇒ A′ x = b′ for x ∈ Rn×1 , that is, the linear
systems Ax = b and A′ x = b′ have the same solutions.
This follows by noting that an interchange of two equations
does not change the solutions, neither does an addition of an
equation to another, nor does a multiplication of an equation
by a nonzero scalar, since these operations can be undone by
similar operations, namely, interchange of the equations in the
reverse order, subtraction of an equation from another, and
division of an equation by a nonzero scalar.
Consequently, we are justified in performing EROs on the
augmented matrix [A|b] in order to obtain all solutions of the
given linear system.

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

Transformation to REF
Let A ∈ Rm×n , that is, let A be an m × n matrix with entries
in R. If A = O, the zero matrix, then it is already in REF.
Suppose A ̸= O.
(i) Let column k1 be the first nonzero column of A, and let
some nonzero entry p1 in this column occur in the jth row of
A. Interchange row j and row 1. Then A is transformed to
 
0 · · · 0 p1 ∗ · · · ∗
0 · · · 0 ∗ ∗ · · · ∗
′
A :=  . . . .. .. ..  ,
 
. . . ..
. . . . . . .
0 ··· 0 ∗ ∗ ··· ∗

where ∗ denotes a real number. Note: p1 becomes the chosen

pivot in row 1. (This choice may not be unique.)
Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03
(ii) Since p1 ̸= 0, add suitable scalar multiples of row 1 of A′
to rows 2 to m of A′ , so that all entries in column k1 below
the pivot p1 are equal to 0. Then A′ is transformed to
 
0 · · · 0 p1 ∗ · · · ∗
0 · · · 0 0 ∗ · · · ∗
′′
A :=  . . . .. .. ..  .
 
. . . ..
. . . . . . .
0 ··· 0 0 ∗ ··· ∗
(iii) Keep row 1 of A′′ intact, and repeat the above process for
the remaining (m − 1) × n submatrix of A′′ to obtain
0 · · · 0 p1 ∗ ∗ ∗ ∗ ··· ∗
 
0 · · · 0 0 · · · 0 p2 ∗ · · · ∗
 
A′′′ := 0 · · · 0 0 · · · 0 0 ∗ · · · ∗ ,
 
. . . .. .. .. .. .. .. .. 
 .. .. .. . . . . . . .
0 ··· 0 0 ··· 0 0 ∗ ··· ∗
′′′
where p2 ̸= 0 and occurs in column k2 of A , where k1 < k2 .
Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03
Note: p2 becomes the chosen pivot in row 2. (Again, this
choice may not be unique.)
(iv) Keep rows 1 and 2 of A′′′ intact, and repeat the above
process till the remaining submatrix has no nonzero row. The
resulting m × n matrix is in REF with pivots p1 , . . . , pr in
columns k1 , . . . , kr , and the last m − r rows are zero rows,
where 1 ≤ r ≤ m.
Notation
Ri ←→ Rj will denote the interchange of the ith row Ri and
the jth row Rj for 1 ≤ i, j ≤ m with i ̸= j.
Ri + αRj will denote the addition of α times the jth row Rj to
the ith row Ri for 1 ≤ i, j ≤ m with i ̸= j.
αRj will denote the multiplication of the jth row Rj by the
nonzero scalar α for 1 ≤ j ≤ m.

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

Remark
A matrix A may be transformed to different
  REFs by EROs.
1 3
For example, we can transform A := by EROs to
    2 0
1 3 2 0
as well as to , both of which are REFs.
0 −6 0 3

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

Examples
(i) Consider the linear system
3x1 + 2x2 + x3 = 3
2x1 + x2 + x3 = 0
6x1 + 2x2 + 4x3 = 6.
We can check that
   
3 2 1 3 3 2 1 3
 2 1 1 0  −→  0 −1/3 1/3 −2 
 
6 2 4 6 0 −2 2 0
 
3 2 1 3
−→  0 −1/3 1/3 −2  .
 
0 0 0 12
Here m = 3 = n, r = 2 and br′ +1 = b3′ = 12 ̸= 0. Hence the
given linear system has no solution.
Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03
(ii) Consider the linear system
x1 − x2 + x3 = 0
−x1 + x2 − x3 = 0
10x2 + 25x3 = 90
20x1 + 10x2 = 80.
As we have already seen in Lecture 2,
   
1 −1 1 0 1 −1 1 0
−1 1 −1 0  EROs  0 10 25 90 
  −−−−→  .
 0 10 25 90 0 0 −95 −190
20 10 0 80 0 0 0 0

Here m = 4, n = 3, r = 3, pivotal columns: 1 , 2 , 3 .

Since br′ +1 = b4′ = 0 and r = n, the linear system has a unique
 T
solution, namely x0 := 2 4 2 , which we had obtained by
back substitution in Lecture 2.
Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03
(iii) Consider the linear system

3.0x1 + 2.0x2 + 2.0x3 − 5.0x4 = 8.0

0.6x1 + 1.5x2 + 1.5x3 − 5.4x4 = 2.7
1.2x1 − 0.3x2 − 0.3x3 + 2.4x4 = 2.1.

We can check that

 
3 2 2 −5 8
[A|b] =  0.6 1.5 1.5 −5.4 2.7 
1.2 −0.3 −0.3 2.4 2.1
 
3 2 2 −5 8
EROs
−−−−−→  0 1.1 1.1 −4.4 1.1  =: [A′ |b′ ].
0 0 0 0 0
Here m = 3, n = 4, r = 2, pivotal columns: 1 , 2 ,
nonpivotal columns: 3 , 4 .
Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03
Since br′ +1 = b3′ = 0, the linear system A′ x = b′ has a solution.
For a particular solution of A′ x = b′ , let x3 = x4 = 0. Then
1.1 x2 = 1.1 =⇒ x2 = 1,
3x1 + 2x2 = 8 =⇒ x1 = 2,
 T
Thus x0 := 2 1 0 0 is a particular solution.
Since r = 2 < 4 = n, the linear system has many solutions.
For basic solutions of A′ x = 0′ , where 0′ = 0, 
 T
let x3 = 1, x4 = 0, so that s3 := 0 −1 1 0 ,
 T
and x4 = 1, x3 = 0, so that s4 := −1 4 0 1 ,
The general solution of A′ x = b′ is given by
x = x0 + α3 s3 + α4 s4 , that is,
x1 = 2 − α4 , x2 = 1 − α3 + 4α4 , x3 = α3 , x4 = α4 , where
α3 , α4 are arbitrary real numbers. These are precisely the
solutions of the given linear system Ax = b.

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

Proposition
Let A ∈ Rm×n . Then the linear system Ax = 0 has only the
zero solution if and only if any REF of A has n nonzero rows.
In particular, if m < n, then Ax = 0 has a nonzero solution.

Proof. We saw that these results hold if A itself is in REF.

Since every A ∈ Rm×n can be transformed to a REF A′ by
EROs, and since the solutions of the linear system Ax = 0 and
the transformed system A′ x = 0′ , where 0′ = 0, are the same,
the desired results hold.
Note: Suppose an m × n matrix A is transformed by EROs to
different REFs A′ and A′′ . Suppose A′ has r ′ nonzero rows
and A′′ has r ′′ nonzero rows. Then 0 ≤ r ′ , r ′′ ≤ min{m, n}.
The above result implies that r ′ = n ⇐⇒ r ′′ = n. We shall
later see that r ′ = r ′′ always.

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

A Challenge Problem
Let A ∈ R9×4 and B ∈ R7×3 . Is there X ∈ R4×7 such that
X ̸= O but AXB = O?

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

Inverse of a Square Matrix
We now introduce a special kind of square matrices.
Let A be a square matrix of size n ∈ N, that is, let A ∈ Rn×n .
We say that A is invertible if there is B ∈ Rn×n such that
AB = I = BA, and in this case, B is called an inverse of A.
Examples  
1 0
The matrix A := is invertible. To see this, let
  0 2
1 0
B := , and check AB = I = BA. On the other hand,
0 1/2  
1 0
the nonzero matrix A := is not invertible. To see this,
  0 0  
a b 2×2 a b
let B := ∈R and note that AB = ̸= I.
c d 0 0

Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03

If A is invertible, then it has a unique inverse. In fact, if
AC = I = BA, then C = IC = (BA)C = B(AC) = BI = B
by the associativity of the matrix multiplication.
If A is invertible, its inverse will be denoted by A−1 , and so
A−1 A = I = AA−1 . Clearly, (A−1 )−1 = A. If A is invertible
and if one can guess its inverse, then it is easy to verify that it
is in fact the inverse of A. Here is a case in point.

Proposition
Let A be a square matrix. Then A is invertible if and only if
AT is invertible. In this case, (AT )−1 = (A−1 )T .

Proof. Suppose A is invertible and B is its inverse. Then

AB = I = BA, and so BT AT = IT = AT BT . Since IT = I, we
see that AT is invertible and (AT )−1 = BT = (A−1 )T .
Next, if AT is invertible, then A = (AT )T is invertible.
Prof. Saurav Bhaumik, IIT Bombay Linear Algebra: Lecture 03